Singleton 2024 — Health Information and the Timing of Social Security Benefit Entitlements

Summary

Exploits random assignment of fasting medical tests in the National Health and Nutrition Examination Survey (NHANES) — morning exams include plasma glucose, low-density lipoprotein (LDL) cholesterol, and triglycerides; afternoon exams do not — to generate exogenous variation in health information among workers aged 50–61. Matched to the Social Security Administration's (SSA) Master Beneficiary Record (1999–2004 surveys, outcomes through 2008), the paper finds that receiving new health information delays Social Security (SS) benefit entitlements by a cumulative 0.380 years for workers aged 59–61 on the eve of the age-62 old-age (OA) eligibility threshold. A parallel increase in covered employment (+9 percentage points (pp)) matches the benefit delay in a near-one-to-one relationship, and the delay dissolves near age 65 when Medicare coverage becomes universal — together identifying employment lock (continued employment to retain employer-sponsored insurance (ESI)) as the dominant mechanism.

Key Claims

• New health information delayed cumulative SS entitlements by 0.380 years for ages 59–61 (sum of event-study coefficients, periods 0–4).
• By period 2, morning examinees at ages 59–61 were 7.79–8.13 pp less likely to be entitled to any SS benefit (mean 65.54%), significant at 5%.
• In period 4, the entitlement gap was 9.46 pp, the only individually significant period estimate.
• Effect is absent at ages 50–58 (no imminent OA eligibility notch).
• Effect dissolves near age 65 (period-5 estimate: 1.8 pp, standard error (SE) = 3.1) consistent with Medicare making ESI redundant.
• Delays attributable to both Disability Insurance (DI) and OA (OA period 1: −5.28 pp, sig. at 5%; period 4: −7.09 pp, sig. at 10%; DI estimates directionally similar but not significant individually).
• Education gradient: −17.10 pp differential for college+ vs. less-educated (significant at 5%); less-educated effect: −1.02 pp (not significant).
• Diabetes mechanism: −10.85 pp for non-previously-diagnosed morning examinees vs. +2.78 pp for previously diagnosed; differential = 13.63 pp (borderline significant).
• Employment confirmation: morning examinees at 59–61 were +11.30 pp more likely to have any SS covered earnings in period 1 and +9.11 pp in period 2 (both significant at 5%).
• ~30% of diabetics in the sample were undiagnosed (consistent with Centers for Disease Control and Prevention (CDC) national estimate of 30.7%).
• Among morning examinees: 41.26% diagnosed with at least one of diabetes/high LDL/high triglycerides; 38.01% newly diagnosed with at least borderline diabetes.
• Scaling: if entire delay is due to new diabetes diagnoses, ~20.5% of newly diagnosed individuals delayed SS claiming.

Concepts Introduced or Extended

• Employment Lock

Entities Mentioned

• Perry Singleton
• Social Security Administration
• Health and Retirement Study

Quotes

"The results suggest that new health information led to delays in entitlements, specifically among individuals who were given new health information just before the earliest entitlement age of 62."

"By period 2, morning examinees were 9.11 percentage points more likely to have any quarters of coverage and were 8.21 percentage points less likely to be entitled to benefits. Combined, the results suggest a roughly one-to-one relationship between a decrease in benefit entitlement and an increase in employment."

My Take

The identification is unusually clean for a health-information study: random assignment is externally confirmed by a National Center for Health Statistics (NCHS) administrator, compliance was near-universal, and the balanced pre-treatment covariates support random assignment. The age-65/Medicare falsification is the paper's strongest evidence — a health-investment mechanism would not predict this dissolve, but employment lock predicts it precisely.

The main limitations are (1) sample size (n = 2,013), yielding wide confidence intervals: the period-2 main estimate spans −0.44 to −15.14 pp; (2) intent-to-treat interpretation, since some afternoon examinees may independently seek fasting tests after total/high-density lipoprotein (HDL) cholesterol alerts; (3) data through 2008 only, so the study predates the Affordable Care Act (ACA), which substantially expanded alternative coverage options for 59–64 year olds. Post-ACA, employment lock from health information may be attenuated for workers with Medicaid or marketplace subsidy access.

The comparison with Benitez-Silva and Dwyer (2005), who find a 0.764-year effect on expected retirement age from a diabetes transition (Health and Retirement Study [HRS] data), is instructive: Singleton's 0.380 actual-claiming estimate is roughly half as large, plausibly because actual claiming responds to insurance institutions while retirement expectations respond to health beliefs directly.